Early Detection of Coronary Artery Disease Using PSO-Based Neuroevolution Model

Abstract

AbstractThere are a large number of causes of morbidness over the world population, and cardiovascular disease is one of the top reasons among them. During the last decade, early detection of warning symptoms is regarded as one of the most important tasks, easing the treatment procedure and preventing a large number of deaths. Numerous people die annually from coronary artery disease (CAD) compared to other deadly diseases. Two main reasons for this disease could be smoking and unhealthy dieting. Angiography is widely used as the most accurate technique for CAD recognition in medical centers; however, some downsides could be mentioned for this technique: It is costly and has various complications such as hematoma and intercostal artery pseudoaneurysms. The machine learning (ML) predictive algorithms bring to medical treatment a new aspect of diagnosing diseases. ML techniques have been applied as powerful and trustworthy tools in the field of medicine for different purposes over the last recent years. ML aims to efficiently extract valuable information from the clinical datasets with minimal effort. In this study, we use a set of machine learning algorithms to reach an accurate prediction of CAD. A powerful nature-inspired optimization algorithm named particle swarm optimization (PSO) based on multilayer perceptron (MLP) and nine states of traditional supervised learning techniques including RF (Random Forest), PSO-MLP, RT (Random Tree), NB (Naïve bayes), SVM (support vector machine), LR (logistic regression), Bagging, MLP, J48 and CART (classification and regression tree analysis) are employed for diagnosis CAD. We implement various machine learning techniques on a straightforward and introduced dataset by the name of Z-Alizadeh Sani. Well-known dataset named Z-Alizadeh Sani to determine the best machine learning technique with the highest accuracy. This dataset contains the information of 303 people who visit the heart disease center. Also, we applied a feature selection method to reduce redundant and outlier features and detect the considerable features that have an important role to play in CAD disease forecasting. The feature selection process improves the power of disease forecasting. Consequently, the obtained accuracy and area under the curve (AUC) from utilized algorithms compared with each other to specify the best technique. This study aims to determine the best model for classifying CAD. Eventually, according to experimental results, the MLP model trained by PSO performs better than all nine art supervised learning algorithms and achieves an accuracy of 90.56% in CAD prediction.KeywordsCoronary artery diseaseCADNeural networkPSONeuroevolution